基于多传感器信息融合的城市边坡监测数据异常事件检测

Abnormal event detection of city slope monitoring data based on multi-sensor information fusion

为预防和管控城市突发地质灾害造成的人民生命和财产损失,国家针对城市地质灾害易发地区部署了大量的各类传感器,用来感知和监测城市边坡等地质体的变化情况,以支持对地质灾害的预警。从边坡监测数据特点和时序数据分析技术出发,针对监测数据噪声混杂、模式分析困难、预警阈值的不确定性等问题,给出了一种基于多传感器信息融合的边坡监测数据异常事件检测方法。主要工作包括:①边坡监测数据变化模式可以归结为周期项、趋势项以及噪声项的叠加,实践中在预处理基础上对边坡监测数据进行周期为24 h的重采样,同时趋势项可以近似看作是经典的牛顿运动,以此构建形变运动模型,为卡尔曼滤波的状态转移提供理论支持;②采用集中式衰减记忆卡尔曼滤波,引入衰减记忆因子,对多传感器边坡监测数据进行特征级融合,降低了噪声的影响,提高了边坡监测数据的可靠性;③引入惩罚系数,应用改进的动态时间弯曲算法对于周期序列数据进行相似性度量。在此基础上基于K-means聚类和局部异常因子分析对边坡监测数据进行异常检测,并基于3σ准则确定预警阈值。该方法能将正常模式和异常模式的时序数据进行区分,有效检测出边坡监测数据的异常,为灾害预防提供支持。最后以深圳市典型边坡监测数据为例验证了此方法的可行性。

To prevent and control the loss of people′s lives and property caused by sudden urban geological disasters,China has deployed a large number of sensors for urban geological disaster-prone areas to perceive changes in urban underground space.In this article,based on the characteristics of slope monitoring data and the analysis technology of time series data,aiming at problems such as noise mixtures in monitoring data,the difficulty of mode analysis and the uncertainty of early warning thresholds,a method of abnormal event detection in slope monitoring data based on multisensor information fusion is proposed.The results show that:①Aiming at the disadvantage that the optimal estimation of the Kalman filter requires known noise information,the attenuation memory factor is introduced,and the centralized attenuation memory Kalman filter is used to fuse the multisensor slope monitoring data,which reduces the influence of noise and improves the reliability of slope monitoring data.②The change mode of slope monitoring data can be summed up as the superposition of periodic term,trend term and noise term.The period is 24 hours,and the trend term can be approximately regarded as the classic Newtonian motion.Based on this,the deformation motion model can be constructed to provide theoretical support for the state transfer of the Kalman filter.③The penalty coefficient is introduced to make the improved DTW have a better measurement effect for the periodic sequence.On this basis,anomaly detection is carried out on the slope monitoring data based on K-means clustering,and local anomaly factors are used to analyse the abnormal conditions of the monitoring data.This method can distinguish the time series data of thenormal mode and abnormal mode better,detect abnormal slope monitoring data effectively,and provide guarantees for disaster prevention.Therefore,in view of the insufficiency of slope monitoring data processing and analysis processes,different information fusion technologies are adopted to improve the reliability and robustness of slope monitoring data.The feasibility of the proposed method is verified by slope monitoring data in Shenzhen.